fix: keep public headers warning-free under -Wextra; release 0.6.1

CHANGELOG.md

@@ -6,6 +6,22 @@ uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+## [0.6.1] - 2026-09-25
+
+### Fixed
+
+- Public headers compile without warnings under `-Wall -Wextra -Wpedantic`,
+  so projects that build with `-Werror` can include them. `Error` values in
+  `error.hpp` and `fixed_point.hpp` left a field uninitialized, which
+  `-Wmissing-field-initializers` reported. The consumer check now builds with
+  those flags.
+- A request refused by the local rate limiter reports `http_status` 0, since
+  no server answered; it reported 429.
+- README's examples copy the `Signer` instead of moving it, so later snippets
+  still have a key, and its Ubuntu setup installs a compiler and a new enough
+  CMake. The docs, comments, and pre-commit hook were corrected where the
+  final review found them out of date.
+
 ## [0.6.0] - 2026-09-25
 
 0.6.0 rebuilds the SDK from Kalshi's published specs. `KalshiClient` covers all
@@ -22,7 +38,7 @@ the 0.5 API; "Migrating from 0.5" below lists what changed.
   document: all 13 channels, `update_subscription` for markets, snapshots, CF
   Benchmarks indices, and Pyth underlyings, and `list_subscriptions`. Message
   types in `kalshi/ws_models.hpp` are generated from `spec/asyncapi.yaml`, and
-  [docs/channels.md](docs/channels.md) maps channels to them.
+  [docs/channels.md](https://github.com/Reddimus/kalshi-cpp/blob/main/docs/channels.md) maps channels to them.
 - The WebSocket client reconnects with backoff after a dropped connection,
   signs each attempt, and resubscribes with each subscription's current
   markets. `ws::Subscription` handles survive reconnects, and every
@@ -40,7 +56,7 @@ the 0.5 API; "Migrating from 0.5" below lists what changed.
   up from 70. New areas include historical data, live data, fee changes, event
   candlesticks and forecasts, order-group triggers and limits, intra-exchange
   transfers, target balance allocation, block trades, API usage levels, search
-  filters, and FCM. [docs/operations.md](docs/operations.md) lists every one.
+  filters, and FCM. [docs/operations.md](https://github.com/Reddimus/kalshi-cpp/blob/main/docs/operations.md) lists every one.
 - `tools/codegen/generate.py` generates the client, models, route tests, and
   operation list from the vendored `spec/openapi.yaml`. CI fails if they drift.
 - Requests are validated before sending: required fields must be set and
@@ -763,7 +779,8 @@ the 0.5 API; "Migrating from 0.5" below lists what changed.
 
 ## [0.0.2] — initial public release
 
-[Unreleased]: https://github.com/Reddimus/kalshi-cpp/compare/v0.6.0...HEAD
+[Unreleased]: https://github.com/Reddimus/kalshi-cpp/compare/v0.6.1...HEAD
+[0.6.1]: https://github.com/Reddimus/kalshi-cpp/compare/v0.6.0...v0.6.1
 [0.6.0]: https://github.com/Reddimus/kalshi-cpp/compare/v0.5.2...v0.6.0
 [0.5.2]: https://github.com/Reddimus/kalshi-cpp/compare/v0.5.1...v0.5.2
 [0.5.1]: https://github.com/Reddimus/kalshi-cpp/compare/v0.5.0...v0.5.1

CLAUDE.md

@@ -8,8 +8,8 @@ User-visible changes also require a `CHANGELOG.md` entry. The CI workflow is
 the source of truth for Linux, macOS, Windows, sanitizer, clang-tidy, Markdown,
 and consumer gates.
 
-Examples make authenticated network calls. Keep automated tests offline by
-using an injected `HttpTransport`.
+Examples call the live API, and most need a key. Keep automated tests offline
+by injecting an `HttpTransport` or using the local servers in `tests/`.
 
 ## Architecture
 
@@ -33,8 +33,9 @@ using an injected `HttpTransport`.
 
 ## Conventions
 
-- Use explicit local types. The permitted `auto` cases are recorded in
-  `tools/cpp_auto_allowlist.txt` and enforced by `tools/cpp_auto_audit.py`.
+- Use explicit local types. `tools/cpp_auto_audit.py` allows `auto` for
+  structured bindings, lambdas, and iterators; anything else needs an
+  `// auto-ok: reason` comment.
 - `tools/codegen/generate.py` writes the REST client from `spec/openapi.yaml`
   and the WebSocket types from `spec/asyncapi.yaml`. Edit a spec, the
   generator, or a `tools/codegen/*.in` template, then run `make codegen`;

CMakeLists.txt

@@ -1,7 +1,7 @@
 # Fetching Glaze needs CMake 3.31; an installed Glaze works with 3.21.
 cmake_minimum_required(VERSION 3.21...4.4)
 project(kalshi-cpp
-    VERSION 0.6.0
+    VERSION 0.6.1
     DESCRIPTION "C++23 client for the Kalshi Predictions API"
     HOMEPAGE_URL "https://github.com/Reddimus/kalshi-cpp"
     LANGUAGES CXX)

CONTRIBUTING.md

@@ -46,11 +46,20 @@ make docs            # Doxygen API reference in build-docs/html (needs Doxygen)
 make bench           # Google Benchmark suite
 make consumers       # Build installed and FetchContent consumers
 make coverage        # lcov report (needs lcov)
+make lint-docs       # markdownlint (needs markdownlint-cli2)
 ```
 
-Run `make format lint test` before pushing. `make install-hooks` runs format
-and lint on every commit. CI adds the sanitizers, clang-tidy, and the consumer
-check, on Linux, macOS, and Windows.
+Run `make format lint test` before pushing, and `make lint-docs` (needs
+`markdownlint-cli2`) after editing Markdown. `make install-hooks` runs
+`make lint` on every commit. CI builds and tests on Linux, macOS, and Windows,
+and runs the sanitizers, clang-tidy, and the consumer check on Linux.
+
+`make tidy` needs clang-tidy 18. On Ubuntu, build with the matching clang and
+libc++, as CI does:
+
+```bash
+make tidy CMAKE_ARGS="-DCMAKE_CXX_COMPILER=clang++-18 -DCMAKE_CXX_FLAGS=-stdlib=libc++"
+```
 
 ## Generated code
 
@@ -90,7 +99,9 @@ explains how to refresh the specs.
 ## Releases
 
 1. Set `VERSION` in `CMakeLists.txt`, move the `[Unreleased]` notes into a new
-   `[X.Y.Z]` section, and update the `GIT_TAG` in `README.md`.
+   `[X.Y.Z]` section, update the compare links at the bottom of
+   `CHANGELOG.md`, and update the `GIT_TAG` in `README.md`. Release notes
+   come from that section, so link with absolute URLs.
 2. Merge that change to `main`.
 3. Run `git tag vX.Y.Z && git push origin vX.Y.Z`. `release.yml` publishes the
    GitHub release after CI passes on the tagged commit, and `docs.yml`

Makefile

@@ -73,10 +73,11 @@ format:
 
 pre-commit: format lint
 
-# Installs a hook that runs `make pre-commit`. Works in worktrees too.
+# Installs a hook that runs `make lint` on every commit, so unformatted code
+# fails instead of being formatted after it is staged. Works in worktrees too.
 install-hooks:
 	@hook="$$(git rev-parse --git-path hooks)/pre-commit"; \
-		printf '#!/bin/sh\nexec make pre-commit\n' > "$$hook" && chmod +x "$$hook" && \
+		printf '#!/bin/sh\nexec make lint\n' > "$$hook" && chmod +x "$$hook" && \
 		echo "Installed $$hook"
 
 # Needs lcov and genhtml.
@@ -113,7 +114,7 @@ help:
 	@echo "make lint-docs    markdownlint"
 	@echo "make format       Format C++ sources in place"
 	@echo "make pre-commit   format + lint"
-	@echo "make install-hooks  Run pre-commit on every git commit"
+	@echo "make install-hooks  Run make lint on every git commit"
 	@echo "make coverage     lcov report in build-coverage/html"
 	@echo "make run-NAME     Run examples/NAME.cpp with .env loaded (ARGS=...)"
 	@echo "make clean        Remove build directories"

README.md

@@ -18,11 +18,14 @@ scope.
 ## Build
 
 You need a C++23 compiler, CMake 3.31+, OpenSSL 3, libcurl, and libwebsockets.
-Ubuntu 24.04's apt CMake is older; `pipx install cmake` gets a current one.
 
 ```bash
-brew install cmake openssl curl libwebsockets pkg-config    # macOS
-sudo apt install cmake libssl-dev libcurl4-openssl-dev libwebsockets-dev pkg-config   # Ubuntu
+# macOS
+brew install cmake openssl curl libwebsockets pkg-config
+
+# Ubuntu 24.04, whose apt CMake is older than 3.31
+sudo apt install build-essential pkg-config pipx libssl-dev libcurl4-openssl-dev libwebsockets-dev
+pipx install cmake && pipx ensurepath   # then open a new shell
 
 cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
 cmake --build build --parallel
@@ -35,7 +38,7 @@ ctest --test-dir build --output-on-failure
 include(FetchContent)
 FetchContent_Declare(kalshi
   GIT_REPOSITORY https://github.com/Reddimus/kalshi-cpp.git
-  GIT_TAG v0.6.0)
+  GIT_TAG v0.6.1)
 FetchContent_MakeAvailable(kalshi)
 target_link_libraries(myapp PRIVATE kalshi::kalshi)
 ```
@@ -73,9 +76,10 @@ int main() {
 
 ## Authenticate
 
-Create a key under API keys at <https://kalshi.com/account/profile>. Kalshi
-recommends Ed25519; generating the pair yourself keeps the private key off the
-website:
+Create a key under API keys at <https://kalshi.com/account/profile> and save
+the private key file Kalshi gives you. Kalshi recommends Ed25519 keys. If the
+page offers to use your own public key, you can generate the pair locally so
+the private key never leaves your machine:
 
 ```bash
 openssl genpkey -algorithm ed25519 -out kalshi.key
@@ -84,7 +88,11 @@ openssl pkey -in kalshi.key -pubout    # paste this public key into Kalshi
 
 ```cpp
 kalshi::Result<kalshi::Signer> signer = kalshi::Signer::from_pem_file(key_id, "kalshi.key");
-kalshi::KalshiClient client{kalshi::HttpClient{std::move(*signer)}};
+if (!signer) {
+    std::cerr << signer.error().message << '\n';
+    return 1;
+}
+kalshi::KalshiClient client{kalshi::HttpClient{*signer}};   // Signer is cheap to copy
 kalshi::Result<kalshi::GetBalanceResponse> balance = client.get_balance();
 ```
 
@@ -121,7 +129,7 @@ Kalshi's status and error code, and `Error::message` includes its explanation.
 Both are transports you stack under the client:
 
 ```cpp
-std::shared_ptr<kalshi::HttpClient> http = std::make_shared<kalshi::HttpClient>(std::move(*signer));
+std::shared_ptr<kalshi::HttpClient> http = std::make_shared<kalshi::HttpClient>(*signer);
 std::shared_ptr<kalshi::RateLimitedTransport> paced =
     std::make_shared<kalshi::RateLimitedTransport>(http, kalshi::RateLimitConfig{});
 kalshi::KalshiClient client{std::make_shared<kalshi::RetryingTransport>(paced)};

include/README.md

@@ -17,6 +17,8 @@
 | `kalshi/error.hpp` | `Error`, `ErrorCode`, and `Result<T>` |
 | `kalshi/environment.hpp` | Production and demo URLs |
 | `kalshi/fixed_point.hpp` | Exact decimal parsing |
+| `kalshi/raw_json.hpp` | `RawJson`, free-form JSON kept as text |
+| `kalshi/version.hpp` | `kalshi::VERSION` (generated at configure time) |
 
 Headers under `kalshi/detail/` support the implementation and tests. They are
 not a stable interface.

include/kalshi/api.hpp

@@ -18,8 +18,9 @@ namespace kalshi {
 /// operation in Kalshi's OpenAPI document, named after its operationId.
 ///
 /// Failures come back as `Error`: `InvalidRequest` for a request the SDK
-/// rejects before sending (or HTTP 400/409/422), `AuthenticationError`,
-/// `NotFound`, `RateLimited`, `ServerError`, `NetworkError`, or `ParseError`.
+/// rejects before sending or another HTTP 4xx, `AuthenticationError`,
+/// `NotFound`, `RateLimited`, `ServerError`, `NetworkError`, `ParseError`,
+/// or `SigningError`.
 /// `Error::api_code` carries Kalshi's machine-readable code when it sends one.
 class KalshiClient {
 public:

include/kalshi/detail/http_path.hpp

@@ -18,7 +18,7 @@ namespace kalshi::detail {
 	return result;
 }
 
-/// Build the URL path covered by Kalshi's RSA-PSS signature.
+/// Build the URL path covered by Kalshi's request signature.
 ///
 /// Kalshi signs the full path from the host root, including the base URL's
 /// `/trade-api/v2` prefix, but excludes the query string.

include/kalshi/error.hpp

@@ -29,19 +29,19 @@ struct Error {
 	std::string api_code;
 
 	[[nodiscard]] static Error network(std::string msg) {
-		return {ErrorCode::NetworkError, std::move(msg)};
+		return {ErrorCode::NetworkError, std::move(msg), 0, {}};
 	}
 
 	[[nodiscard]] static Error auth(std::string msg) {
-		return {ErrorCode::AuthenticationError, std::move(msg)};
+		return {ErrorCode::AuthenticationError, std::move(msg), 0, {}};
 	}
 
 	[[nodiscard]] static Error parse(std::string msg) {
-		return {ErrorCode::ParseError, std::move(msg)};
+		return {ErrorCode::ParseError, std::move(msg), 0, {}};
 	}
 
 	[[nodiscard]] static Error signing(std::string msg) {
-		return {ErrorCode::SigningError, std::move(msg)};
+		return {ErrorCode::SigningError, std::move(msg), 0, {}};
 	}
 };
 

include/kalshi/fixed_point.hpp

@@ -14,13 +14,14 @@ class FixedPoint {
 public:
 	[[nodiscard]] static Result<FixedPoint> parse(std::string_view value) {
 		if (value.empty()) {
-			return std::unexpected(Error{ErrorCode::InvalidRequest, "fixed-point value is empty"});
+			return std::unexpected(
+				Error{ErrorCode::InvalidRequest, "fixed-point value is empty", 0, {}});
 		}
 
 		std::size_t index = value.front() == '-' ? 1 : 0;
 		if (index == value.size()) {
 			return std::unexpected(
-				Error{ErrorCode::InvalidRequest, "fixed-point value has no digits"});
+				Error{ErrorCode::InvalidRequest, "fixed-point value has no digits", 0, {}});
 		}
 
 		bool saw_digit = false;
@@ -37,10 +38,12 @@ class FixedPoint {
 				saw_decimal = true;
 				continue;
 			}
-			return std::unexpected(Error{ErrorCode::InvalidRequest, "invalid fixed-point value"});
+			return std::unexpected(
+				Error{ErrorCode::InvalidRequest, "invalid fixed-point value", 0, {}});
 		}
 		if (!saw_digit || (saw_decimal && !fractional_digit)) {
-			return std::unexpected(Error{ErrorCode::InvalidRequest, "invalid fixed-point value"});
+			return std::unexpected(
+				Error{ErrorCode::InvalidRequest, "invalid fixed-point value", 0, {}});
 		}
 		return FixedPoint(std::string(value));
 	}
@@ -51,8 +54,8 @@ class FixedPoint {
 	/// Returns InvalidRequest instead of rounding or saturating.
 	[[nodiscard]] Result<std::int64_t> scaled_integer(std::uint8_t scale) const {
 		if (scale > 18) {
-			return std::unexpected(
-				Error{ErrorCode::InvalidRequest, "fixed-point scale exceeds int64 capacity"});
+			return std::unexpected(Error{
+				ErrorCode::InvalidRequest, "fixed-point scale exceeds int64 capacity", 0, {}});
 		}
 
 		const bool negative = wire_.front() == '-';
@@ -67,7 +70,9 @@ class FixedPoint {
 			for (std::size_t i = fractional_start + scale; i < wire_.size(); ++i) {
 				if (wire_[i] != '0') {
 					return std::unexpected(Error{ErrorCode::InvalidRequest,
-												 "fixed-point conversion would lose precision"});
+												 "fixed-point conversion would lose precision",
+												 0,
+												 {}});
 				}
 			}
 		}
@@ -109,7 +114,7 @@ class FixedPoint {
 
 	[[nodiscard]] static Result<std::int64_t> overflow_error() {
 		return std::unexpected(
-			Error{ErrorCode::InvalidRequest, "fixed-point value does not fit in int64"});
+			Error{ErrorCode::InvalidRequest, "fixed-point value does not fit in int64", 0, {}});
 	}
 
 	std::string wire_;

include/kalshi/version.hpp.in

@@ -1,28 +1,19 @@
 #pragma once
 
 /// @file version.hpp
-/// @brief Version constants for the Kalshi C++ SDK.
-///
-/// This header is generated at CMake configure time from
-/// `version.hpp.in` via `configure_file()`. The substituted values
-/// come from CMakeLists.txt's `project(kalshi-cpp VERSION ...)`
-/// declaration — that's the single source of truth.
-///
-/// Do NOT edit the generated copy in the build directory. Edit
-/// CMakeLists.txt to change the version; this header re-generates.
+/// @brief The SDK version, generated from `project(... VERSION)` in CMakeLists.txt.
 
 namespace kalshi {
 
 /// Full SDK version, "MAJOR.MINOR.PATCH".
 constexpr const char* VERSION = "@PROJECT_VERSION@";
 
-/// Major version component (incompatible API changes).
 constexpr int VERSION_MAJOR = @PROJECT_VERSION_MAJOR@;
 
-/// Minor version component (backwards-compatible additions).
+/// While VERSION_MAJOR is 0, a new minor version may break the API.
 constexpr int VERSION_MINOR = @PROJECT_VERSION_MINOR@;
 
-/// Patch version component (backwards-compatible fixes).
+/// Patch releases fix bugs without changing the API.
 constexpr int VERSION_PATCH = @PROJECT_VERSION_PATCH@;
 
 } // namespace kalshi

src/http/rate_limited_transport.cpp

@@ -117,8 +117,9 @@ Result<HttpResponse> RateLimitedTransport::request(HttpMethod method, std::strin
 				std::to_string(bucket.config().capacity) + "; split the batch"});
 	}
 	if (!bucket.acquire_for(tokens, config_.max_wait)) {
-		return std::unexpected(Error{ErrorCode::RateLimited,
-									 "Rate-limit tokens would not refill within max_wait", 429});
+		// No server answered, so http_status stays 0.
+		return std::unexpected(Error{
+			ErrorCode::RateLimited, "Rate-limit tokens would not refill within max_wait", 0, {}});
 	}
 	return inner_->request(method, path, body);
 }

tests/README.md

@@ -13,6 +13,11 @@ memory.
 | `test_http_client.cpp` | The libcurl transport, POSIX only |
 | `test_transports.cpp` | Retries, token buckets, rate limiting |
 | `test_signer.cpp` | Ed25519 and RSA-PSS signing |
+| `test_helpers.cpp` | Cents, contracts, timestamps, order directions, pagination |
+| `test_fixed_point.cpp` | Exact decimal parsing and scaling |
+| `test_features.cpp` | Defaults for configs and enums |
+| `test_version.cpp` | `kalshi::VERSION` and its components |
+| `parse_benchmark.cpp` | A coarse throughput guard, run as its own test |
 | `test_ws_messages.cpp` | Every example frame in the AsyncAPI spec (generated) |
 | `test_ws_frames.cpp` | Control frames, discriminated types, nulls, unknown values |
 | `test_ws_subscriptions.cpp` | Command frames, held commands, resubscribing, gaps |

tests/parse_benchmark.cpp

@@ -62,7 +62,7 @@ int main(int argc, char** argv) {
 	}
 
 	if (glaze_checksum == 0) {
-		std::fprintf(stderr, "checksum is zero — render_body emitted nothing\n");
+		std::fprintf(stderr, "checksum is zero: encode() produced nothing\n");
 		return 1;
 	}
 
@@ -73,10 +73,7 @@ int main(int argc, char** argv) {
 	std::printf("parse_benchmark: payload=%zuB iters=%d\n", sample.size(), kIterations);
 	std::printf("  glaze (serialize): %8.3f ms total  (%8.3f us/op)\n", glaze_ms, us_per_op);
 
-	// Regression guard: at migration time, Glaze rendered a 50-order
-	// batch in ~30-60 us/op on x86_64-v3 / -O3 / LTO. Cap at 500 us/op
-	// — that's ~10x the measured baseline and accounts for slower CI
-	// runners, Debug builds, and AddressSanitizer overhead.
+	// About 10x the measured cost of encoding a 50-order batch, to absorb slow runners.
 	constexpr double kMaxUsPerOp = 500.0;
 	if (check_timing && us_per_op > kMaxUsPerOp) {
 		std::fprintf(stderr, "REGRESSION: %.3f us/op exceeds cap of %.0f us/op\n", us_per_op,